Credit Scoring Solution Applied Methodology for Credit Insurance Juanjo Ortiz Osorio Risk Analysis Programme Manager SAS Spain

Transcription

1 Credit Scoring Solution Applied Methodology for Credit Insurance Juanjo Ortiz Osorio Risk Analysis Programme Manager SAS Spain Copyright 2004, SAS Institute Inc. All rights reserved. 17 June 2004

2 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

3 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

4 The Company About Crédito y Caución (CyC): It s the most important insurance company in Spain. It has the 80% of the whole domestic market. The company also works in Andorra and Portugal. Relationship with SAS. SAS Enterprise Miner. Education. First steps in ETL. First statistical analysis.

5 Introduction SAS Credit Scoring solution allows you to: Develop scoring models to assign probabilities of default to the counterparties (retail, corporations, etc.) Credit scorecard development: automatic and interactive grouping of variables, logistic regression and a score rating. Scorecard evaluation reports. Decision Trees. HTML reporting. Develop LGD models. Define rating scale.

6 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

7 Credit Scorecards Goal: Get the score for each of the variables characteristics. The higher the score the lower the risk. The number of points assigned to each characteristic depends on two factors: WOE (weight of evidence): characteristic s contribution to the score of the variable. IV (information value): variable s contribution to the total score. Statistically: Scorecard = Previous study + Logistic Regression Interactive Grouping Using WOE s The total score will be the sum of each score of each variable characteristic

8 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

9 Credit Risk Scorecard Development. Sampling Goal: An insurance company wants to obtain the probability of default for each of their customers. Default (bad): unpayment in the next 12 months. Information about Customers Observation date Behavioural Date Observation period 2 years Behavioural period 1 year Risk DM Unique record for each corporation: netted and summarized information Default ID Sample

10 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

11 Credit Risk Scorecard Development. Data preparation Explore Data Choose the role of each variable. Assign weight to the target variable. Filter Outliers Key step: be careful not to clean too many bads and too many records. Riesgo Total último trimestre

12 Credit Risk Scorecard Development. Data preparation Transform variables Mathematical transformations. Create new variables combining some of them. Data Partition Training data set (e.g. 60% of the sample). Validation data set (e.g. 40% of the sample). Stratified partition in order to maintain the proportion of bads in the samples. Interactive grouping of the variables Select the most predictive variables. Increase predictivity, introducing business knowledge. Grouping criteria: WOE curve with smooth trend IV as big as possible

13 Credit Risk Scorecard Development. Data preparation Interactive grouping of the variables The lower the WOE the higher the risk

14 Credit Risk Scorecard Development. Data preparation Interactive grouping of the variables. Grouping modification clic

15 Credit Risk Scorecard Development. Data preparation Interactive grouping of the variables. Grouping modification clic clic

16 Credit Risk Scorecard Development. Data preparation Interactive grouping of the variables. Results IV 0.02, NON predictive 0.02 < IV 0.1, low predictive 0.1 < IV 0.3, medium predictive 0.3 < IV 0.5, high predictive IV > 0.5, overpredictive clic

17 Credit Risk Scorecard Development. Summary part I

18 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

19 Credit Risk Scorecard Development. Modelling Model development. First model Models developed with the variables obtained in the grouping process: Group variable: called variable_grp and contains the group to which each individual belongs, Label variable: called variable_lbl and contains the description of the group to which each individual belongs, WOE variable: called variable_woe and contains the WOE s of the group. This variable will be used in the model. Reg 1 Reg 2 Reg 3 Reg 4 Assessment Reg 5 Reg 6

20 Credit Risk Scorecard Development. Modelling Model development. First result

21 Credit Risk Scorecard Development. Modelling Model development. Final model Goal: Obtain a high quality final model (optimum power) which supplies the maximum possible business knowledge. Model combination.

22 Credit Risk Scorecard Development. Summary part II Reg 1 Reg 2 Reg 3 Reg 4 Assessment Reg 5 Reg 6 Combinación 1 Control Point Combinación 2 Assessment Combinación 3

23 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

24 Credit Risk Scorecard Development. Reporting Score Analysis Graphs and charts to determine the optimum cut-off point. Ability to score the population with the final model. Score Node Allows use of the model code to score the data set. It must be used before the Score Analysis node.

25 Credit Risk Scorecard Development. Reporting Score Analysis

26 Credit Risk Scorecard Development. FINAL flow diagram Reg 1 Reg 2 Reg 3 Reg 4 Assessment Reg 5 Reg 6 Combinación 1 Control Point Combinación 2 Assessment MODELO ELEGIDO Score Combinación 3 Score Analysis

27 Credit Risk Scorecard Development. Reporting Scorecard

28 Credit Risk Scorecard Development. Reporting Score distribution

29 Credit Risk Scorecard Development. Reporting

30 Credit Risk Scorecard Development. Reporting % and the approval rate increases % New rules improve the bad rate Tasa de malos vs. Tasa de aprobación 1974

31 Credit Risk Scorecard Development. Reporting Score distribution: goods vs. bads Over 1974 points we authorize most goods and some bads 1974

32 Agenda Introduction Credit Scorecards Credit Risk Scorecard Development Sampling Data preparation Modelling Reporting Monitoring

33 Credit Risk Scorecard Development. Monitoring Month 12 Month 1

34 The Power to Know